nitin.palleti

Open to Software Engineering & Applied AI roles

Nitin Palleti

Senior software engineer with 8+ years on banking, payments and commerce platforms — now building AI systems you can audit: facts first, model second, and an evaluation gate before anything ships.

LinkedIn ↗ GitHub ↗ Résumé ↓ Get in touch
[1] Claims on this page cite their evidence. Hover or tap the brackets.
01

The 30-second brief

Now
Senior Software Engineer at MAS Global Consulting (client: JPMorganChase, Offers team). Built and shipped Agentic Compass, an AI agent for GitHub Copilot, to production.
Track record
8+ years building and running production systems for JPMorganChase, BNY, Copart and Sunrise across banking, payments, telecom and commerce
AI I've shipped
Agentic Compass (in production at work) · Ask Nitin (live on this site) · PaymentOps Workbench (eval-gated AI review)
Strengths
Java microservices on AWS · agentic AI with guardrails · evals as release gates · owning deploys, monitoring and incidents
Looking for
Senior Software Engineer, Applied AI Engineer, Forward Deployed Engineer or AI Platform roles
Stack
Java · Spring Boot · AWS (ECS, S3, EMR) · Kafka · Python · FastAPI · Claude & Copilot agents · MCP · React · PostgreSQL · Docker
Location
Dallas–Fort Worth, TX · Hybrid or remote

What I bring to an applied AI team

  • An AI agent taken all the way to production in regulated finance: senior-lead reviews → UAT → production [7]
  • Guardrails by design: human-approved writes, least-privilege routing, append-only audit trails [7]
  • Evals as release gates: an LLM-judged test suite with a deploy threshold and a held-out set [8]
  • 8+ years owning production services: deployments, monitoring, incident triage [6]
  • Grounded answers over confident guesses: citations, evidence labels, documented limits [5]
02

Featured work

Work project · In production Python · JavaScript · GitHub Copilot — 2026

Agentic Compass

A human-supervised AI agent for GitHub Copilot that turns a Jira ticket into business context, evidence-backed analysis, a routed implementation plan, real test verification and an auditable history, with every remote write approved by a human. Built in my current role at MAS Global Consulting (client: JPMorganChase); in production after senior engineering lead reviews and UAT.

Read the case study →
8real tickets it was validated on
5services across Java and Python
73unit tests, all passing
19routes to existing specialist AI skills

Python · FastAPI · Claude — 2026

Ask Nitin

The assistant in the corner of this page. Ask about my experience and every answer shows where it came from, or paste a job description to see how I match it, requirement by requirement. It answers only from my résumé and project notes, and says so when something isn't covered.

40/42graded test cases passed, checked by rules and an AI judge · Nov 13, 2026
1.9 saverage time to a full chat answer
5.5 saverage time for a job-fit report
$0.0003average model cost per question

How it's built

  1. Find the facts first. A keyword search (BM25) picks the relevant sections of my résumé and project notes; the model answers only from those and cites each one.
  2. Check the model's work. Claude Haiku answers questions; Claude Sonnet writes fit reports, and plain code re-checks every row before you see it.
  3. Safe to leave public. Off-topic requests and prompt injection are refused, with per-visitor rate limits and a daily spending cap.
  4. Measured, not assumed. A test suite with an AI judge and a held-out set it was never tuned on has to pass before deploying.

Pass rate and times: evaluation run of 43 cases on Nov 13, 2026 (one case went ungraded when the run hit its budget; three job-fit cases were re-run once). A separate held-out set passed 10/10. Cost: the average over 491 questions in the request log where a model was actually called (median $0.00023).

Python · Java · TypeScript — 2026

PaymentOps Workbench

A synthetic payment-investigation workbench where every AI suggestion is grounded in cited evidence and gated by a reusable evaluation harness. [1]

View source ↗
“Payment investigation is an evidence problem, not a text-generation problem.”

Architecture — click a component

10synthetic cases with ordered timelines
17service tests passing (10 Spring · 7 FastAPI)
30/20/10dev / held-out / adversarial eval split
0browser-supplied evidence reaching the model

Decisions I'd defend in an interview

  1. Facts before AI. Timelines, flags, policy eligibility and citations are deterministic and visible without a model.
  2. Fail closed. Unknown source IDs, malformed output, deadlines and rate limits become explicit review-required states — never fake success. [2]
  3. Hold out the holdout. The held-out set is fingerprinted, not consumed, until an independent adapter is ready. [3]

Known limitations by design

  • Synthetic data only — no payment rails, no money movement.
  • Tenant selection is caller-controlled; it is not auth.
  • Vector retrieval uses an offline fixture, not an embedding-quality claim.
  • Bedrock adapter and public demo are gated on access and review. [5]
03

How I approach applied AI

deterministic →

Make the facts boring

Retrieve, order and validate the evidence in plain code first. The model should summarise facts, not invent them.

contract →

Typed boundaries around the model

Schema-validated requests and responses, source-ID checks, timeouts and budgets. A model is a dependency that can fail like any other. [2]

eval →

Measure before you promote

Versioned dev, held-out and adversarial sets with a gate that blocks release when the evidence isn't there. [3]

human →

Keep a human in the loop

Drafts are versioned, approvals use optimistic concurrency and leave one durable audit event. [4]

Try it: a tiny eval gate

Illustrative example — click a model response to see whether it would pass.

Select a response ↑
04

Experience

  1. Oct 2024 — Present
    Plano, TX · Hybrid

    Senior Software Engineer · MAS Global Consulting (client: JPMorganChase, Offers team)

    • Design and evolve Spring Boot microservices and REST APIs for personalised offer decisioning across transactional, merchant and behavioural data.
    • Run containerised services on AWS ECS with S3, EMR and IAM; own production deployments for application and infrastructure changes, including a migration from Terraform to EAC.
    • Built offer experimentation on Split feature flags — phased rollouts, controlled exposure and rollback.
    • Designed and built Agentic Compass, a human-supervised AI agent for GitHub Copilot, now in production after senior engineering lead reviews and UAT.
    • Use and extend internal AI skills, MCP tools and Copilot for code analysis, documentation and unit-test modernisation; built AI-assisted workflows that make UAT data preparation repeatable.
    • BlazeMeter performance testing for release readiness; Datadog APM monitors and alerts, plus Splunk and Dynatrace for production triage.
  2. 2026
    Independent

    PaymentOps Workbench · Personal project

    Designed and built a three-service system (React, Spring Boot, FastAPI) with evidence retrieval, an approval workflow and a reusable AI Eval Gate.

  3. Aug 2024 — Sep 2024
    Frisco, TX · Remote

    Java Software Engineer · BNY

    Enhanced Spring Boot services for data ingestion, validation and processing on the Bridgewater platform; built REST integrations and Spring Data / JDBC access layers and optimised queries.

  4. Feb 2024 — Jul 2024
    Plano, TX · Hybrid

    Java Software Engineer · JPMorgan Chase & Co.

    Built Spring Boot microservices for real-time payment workflows with Hibernate/JPA and Spring Security; JMS messaging to the orchestration engine; Dockerised deployments on AWS.

  5. May 2022 — Jan 2024
    Dallas, TX · Hybrid

    Java Full Stack Engineer · Copart

    Microservices and REST APIs for auction, inventory and partner-integration platforms; event-driven messaging with Kafka and ActiveMQ; reusable React components; Oracle and Cassandra on AWS.

  6. Aug 2017 — Mar 2021
    Bengaluru, India

    Java/J2EE Developer · Sunrise (Switzerland)

    Enterprise applications for telecom provisioning and service activation using Spring, Hibernate, REST, JMS and Kafka; production support across the SDLC.

Education

  1. Aug 2021 — May 2023

    M.S. Computer Science · The University of Texas at Arlington

  2. Aug 2013 — May 2017

    Bachelor of Engineering · Amrita Vishwa Vidyapeetham

Toolbox

Backend
Java, Spring Boot, Spring MVC, Spring Data JPA, Hibernate, Spring Security, REST, microservices
AI / LLM
FastAPI AI boundaries, evaluation harnesses, retrieval with pgvector, MCP tools, Copilot
Cloud & DevOps
AWS (ECS, S3, EMR, IAM, EC2, RDS), Docker, Terraform, EAC, Jenkins, CI/CD
Messaging & data
Kafka, ActiveMQ, JMS, PostgreSQL, Oracle, Cassandra, Snowflake
Quality & ops
JUnit, BlazeMeter, Split feature flags, Datadog APM, Splunk, Dynatrace
Frontend
React, JavaScript, HTML5, CSS3
05

Let's talk

Hiring for software engineering or applied AI? I'd like to hear about it.

LinkedInin/nitinpalleti ↗ GitHubnitinpalleti-swe ↗ Emailnpalleti96@gmail.com RésuméDownload PDF ↓